2

I want to compare the effect of a drug (immunomodulator) on antibody levels in a group of patients $(n=18)$. Blood samples were taken pre- and post-treatment in each patient and the means of antibody levels compared for each study subject. At this point I guess that the best and simplest option is a paired t-test.

However, in some cases, I have a different number of blood samples in the pre- vs. the post-treatment group. For example, for one patient I have $16$ samples pre-treatment and $5$ post-treatment.

  • Can I still use the paired t-test?
  • Should I use the independent t-test?
amoeba
  • 93,463
  • 28
  • 275
  • 317
user43181
  • 21
  • 1
  • For each of the measurements, do you know the ordering? Ie, do you know which of the 16 pre-treatment blood samples was drawn first & when, which 2nd & when, etc? Then all measurements could be placed in time. – gung - Reinstate Monica Apr 06 '14 at 03:36
  • yes I have the date of each blood sample – user43181 Apr 06 '14 at 03:37
  • Thanks for fast response! yes I have the date of each blood sample... including the pre and post treatment measures. So they can be ordered by time. But I am not sure if the paired T test is the ideal test to use? When I use JMP v9 and do "matched pairs" the program just compare the means of the first 5 pre treatment values (instead of taking the mean from the whole 16 values) vs the mean of the the 5 post treatment values... Also I read that you need the exact number of measures pre and post to perform the paired t test? So I don't know if I can do the independent T test instead? – user43181 Apr 06 '14 at 03:53
  • 1
    The simplest approach is to take as pairs the means of the pre-treatment measurements and the means of the post-treatment measurements, and do a paired t-test on those means. You should *not* do an independent t-test because the hypothesis you are testing is whether the treatment makes a difference, and the variation between subjects is not accounted for under such a test. However, even this simple approach is not ideal: you have captured variability in the repeated measures. Something like ANOVA would be more appropriate. – heropup Apr 06 '14 at 03:55
  • Thank you so much heropup! How can I manage the fact that I have 16 measures pre-treatment vs 5 measures post-treatment? Does this affect in some way the result? – user43181 Apr 06 '14 at 04:01
  • How do those 16 probes differ? Can you lose some important information if you average them? Likewise, how do those 5 probes differ? If they are 1 day, 2 days,... later the drug admission then these may reflect the temporal development of the effect, which is potentially important. – ttnphns Apr 06 '14 at 06:03

0 Answers0